Cursor is an AI-first code editor forked from VS Code, built by Anysphere. It embeds large language models into the core editing experience so that developers can write, refactor, and review code through natural-language conversation and inline completions.
Key capabilities
Chat and Composer — Cursor’s two main modes. Chat is a side-panel conversation that can read and edit the current file. Composer (Ctrl+K / Cmd+K) opens a full multi-file agent mode that can create and edit files across an entire project in a single prompt.
Codebase indexing — Cursor indexes the entire repository into a vector store so that context-window limits do not prevent it from reasoning across large projects. Queries retrieve semantically relevant files before the LLM sees them.
Model flexibility — Users can route requests to Claude 3.5 Sonnet, GPT-4o, Gemini, or Cursor’s own fine-tuned cursor-small model. The choice is per-request, allowing cost/quality trade-offs.
Tab completion — A speculative autocomplete that predicts not just the next token but entire diff hunks, trained on accept/reject signals from millions of developers.
Autonomy level
Cursor sits at level 3 (supervised agent): it proposes changes and applies them on explicit approval. The Composer mode can chain multiple file edits in sequence, but the developer reviews and accepts each checkpoint. It does not run autonomously in the background or deploy code without human confirmation.
Strengths
- Familiar VS Code environment lowers adoption friction
- Multi-file Composer edits are well-suited to refactoring and feature addition
- Codebase indexing makes it practical on large repos
Limitations
- Closed source; pricing model may change
- Composer can produce large diffs that are hard to review all at once
- Quality depends heavily on the underlying model chosen